dlmcommon.h

linux2.6.16版本

/* -*- mode: c; c-basic-offset: 8; -*-
 * vim: noexpandtab sw=8 ts=8 sts=0:
 *
 * dlmcommon.h
 *
 * Copyright (C) 2004 Oracle.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License as published by the Free Software Foundation; either
 * version 2 of the License, or (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 *
 */

#ifndef DLMCOMMON_H
#define DLMCOMMON_H

#include <linux/kref.h>

#define DLM_HB_NODE_DOWN_PRI     (0xf000000)
#define DLM_HB_NODE_UP_PRI       (0x8000000)

#define DLM_LOCKID_NAME_MAX    32

#define DLM_DOMAIN_NAME_MAX_LEN    255
#define DLM_LOCK_RES_OWNER_UNKNOWN     O2NM_MAX_NODES
#define DLM_THREAD_SHUFFLE_INTERVAL    5     // flush everything every 5 passes
#define DLM_THREAD_MS                  200   // flush at least every 200 ms

#define DLM_HASH_BUCKETS     (PAGE_SIZE / sizeof(struct hlist_head))

enum dlm_ast_type {
	DLM_AST = 0,
	DLM_BAST,
	DLM_ASTUNLOCK
};


#define LKM_VALID_FLAGS (LKM_VALBLK | LKM_CONVERT | LKM_UNLOCK | \
			 LKM_CANCEL | LKM_INVVALBLK | LKM_FORCE | \
			 LKM_RECOVERY | LKM_LOCAL | LKM_NOQUEUE)

#define DLM_RECOVERY_LOCK_NAME       "$RECOVERY"
#define DLM_RECOVERY_LOCK_NAME_LEN   9

static inline int dlm_is_recovery_lock(const char *lock_name, int name_len)
{
	if (name_len == DLM_RECOVERY_LOCK_NAME_LEN &&
	    memcmp(lock_name, DLM_RECOVERY_LOCK_NAME, name_len)==0)
		return 1;
	return 0;
}

#define DLM_RECO_STATE_ACTIVE  0x0001

struct dlm_recovery_ctxt
{
	struct list_head resources;
	struct list_head received;
	struct list_head node_data;
	u8  new_master;
	u8  dead_node;
	u16 state;
	unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
	wait_queue_head_t event;
};

enum dlm_ctxt_state {
	DLM_CTXT_NEW = 0,
	DLM_CTXT_JOINED,
	DLM_CTXT_IN_SHUTDOWN,
	DLM_CTXT_LEAVING,
};

struct dlm_ctxt
{
	struct list_head list;
	struct hlist_head *lockres_hash;
	struct list_head dirty_list;
	struct list_head purge_list;
	struct list_head pending_asts;
	struct list_head pending_basts;
	unsigned int purge_count;
	spinlock_t spinlock;
	spinlock_t ast_lock;
	char *name;
	u8 node_num;
	u32 key;
	u8  joining_node;
	wait_queue_head_t dlm_join_events;
	unsigned long live_nodes_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
	unsigned long domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
	unsigned long recovery_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
	struct dlm_recovery_ctxt reco;
	spinlock_t master_lock;
	struct list_head master_list;
	struct list_head mle_hb_events;

	/* these give a really vague idea of the system load */
	atomic_t local_resources;
	atomic_t remote_resources;
	atomic_t unknown_resources;

	/* NOTE: Next three are protected by dlm_domain_lock */
	struct kref dlm_refs;
	enum dlm_ctxt_state dlm_state;
	unsigned int num_joins;

	struct o2hb_callback_func dlm_hb_up;
	struct o2hb_callback_func dlm_hb_down;
	struct task_struct *dlm_thread_task;
	struct task_struct *dlm_reco_thread_task;
	wait_queue_head_t dlm_thread_wq;
	wait_queue_head_t dlm_reco_thread_wq;
	wait_queue_head_t ast_wq;
	wait_queue_head_t migration_wq;

	struct work_struct dispatched_work;
	struct list_head work_list;
	spinlock_t work_lock;
	struct list_head dlm_domain_handlers;
	struct list_head	dlm_eviction_callbacks;
};

/* these keventd work queue items are for less-frequently
 * called functions that cannot be directly called from the
 * net message handlers for some reason, usually because
 * they need to send net messages of their own. */
void dlm_dispatch_work(void *data);

struct dlm_lock_resource;
struct dlm_work_item;

typedef void (dlm_workfunc_t)(struct dlm_work_item *, void *);

struct dlm_request_all_locks_priv
{
	u8 reco_master;
	u8 dead_node;
};

struct dlm_mig_lockres_priv
{
	struct dlm_lock_resource *lockres;
	u8 real_master;
};

struct dlm_assert_master_priv
{
	struct dlm_lock_resource *lockres;
	u8 request_from;
	u32 flags;
	unsigned ignore_higher:1;
};


struct dlm_work_item
{
	struct list_head list;
	dlm_workfunc_t *func;
	struct dlm_ctxt *dlm;
	void *data;
	union {
		struct dlm_request_all_locks_priv ral;
		struct dlm_mig_lockres_priv ml;
		struct dlm_assert_master_priv am;
	} u;
};

static inline void dlm_init_work_item(struct dlm_ctxt *dlm,
				      struct dlm_work_item *i,
				      dlm_workfunc_t *f, void *data)
{
	memset(i, 0, sizeof(*i));
	i->func = f;
	INIT_LIST_HEAD(&i->list);
	i->data = data;
	i->dlm = dlm;  /* must have already done a dlm_grab on this! */
}



static inline void __dlm_set_joining_node(struct dlm_ctxt *dlm,
					  u8 node)
{
	assert_spin_locked(&dlm->spinlock);

	dlm->joining_node = node;
	wake_up(&dlm->dlm_join_events);
}

#define DLM_LOCK_RES_UNINITED             0x00000001
#define DLM_LOCK_RES_RECOVERING           0x00000002
#define DLM_LOCK_RES_READY                0x00000004
#define DLM_LOCK_RES_DIRTY                0x00000008
#define DLM_LOCK_RES_IN_PROGRESS          0x00000010
#define DLM_LOCK_RES_MIGRATING            0x00000020

/* max milliseconds to wait to sync up a network failure with a node death */
#define DLM_NODE_DEATH_WAIT_MAX (5 * 1000)

#define DLM_PURGE_INTERVAL_MS   (8 * 1000)

struct dlm_lock_resource
{
	/* WARNING: Please see the comment in dlm_init_lockres before
	 * adding fields here. */
	struct hlist_node hash_node;
	struct kref      refs;

	/* please keep these next 3 in this order
	 * some funcs want to iterate over all lists */
	struct list_head granted;
	struct list_head converting;
	struct list_head blocked;

	struct list_head dirty;
	struct list_head recovering; // dlm_recovery_ctxt.resources list

	/* unused lock resources have their last_used stamped and are
	 * put on a list for the dlm thread to run. */
	struct list_head purge;
	unsigned long    last_used;

	unsigned migration_pending:1;
	atomic_t asts_reserved;
	spinlock_t spinlock;
	wait_queue_head_t wq;
	u8  owner;              //node which owns the lock resource, or unknown
	u16 state;
	struct qstr lockname;
	char lvb[DLM_LVB_LEN];
};

struct dlm_migratable_lock
{
	__be64 cookie;

	/* these 3 are just padding for the in-memory structure, but
	 * list and flags are actually used when sent over the wire */
	__be16 pad1;
	u8 list;  // 0=granted, 1=converting, 2=blocked
	u8 flags;

	s8 type;
	s8 convert_type;
	s8 highest_blocked;
	u8 node;
};  // 16 bytes

struct dlm_lock
{
	struct dlm_migratable_lock ml;

	struct list_head list;
	struct list_head ast_list;
	struct list_head bast_list;
	struct dlm_lock_resource *lockres;
	spinlock_t spinlock;
	struct kref lock_refs;

	// ast and bast must be callable while holding a spinlock!
	dlm_astlockfunc_t *ast;
	dlm_bastlockfunc_t *bast;
	void *astdata;
	struct dlm_lockstatus *lksb;
	unsigned ast_pending:1,
		 bast_pending:1,
		 convert_pending:1,
		 lock_pending:1,
		 cancel_pending:1,
		 unlock_pending:1,
		 lksb_kernel_allocated:1;
};


#define DLM_LKSB_UNUSED1           0x01
#define DLM_LKSB_PUT_LVB           0x02
#define DLM_LKSB_GET_LVB           0x04
#define DLM_LKSB_UNUSED2           0x08
#define DLM_LKSB_UNUSED3           0x10
#define DLM_LKSB_UNUSED4           0x20
#define DLM_LKSB_UNUSED5           0x40
#define DLM_LKSB_UNUSED6           0x80


enum dlm_lockres_list {
	DLM_GRANTED_LIST = 0,
	DLM_CONVERTING_LIST,
	DLM_BLOCKED_LIST
};

static inline struct list_head *
dlm_list_idx_to_ptr(struct dlm_lock_resource *res, enum dlm_lockres_list idx)
{
	struct list_head *ret = NULL;
	if (idx == DLM_GRANTED_LIST)
		ret = &res->granted;
	else if (idx == DLM_CONVERTING_LIST)
		ret = &res->converting;
	else if (idx == DLM_BLOCKED_LIST)
		ret = &res->blocked;
	else
		BUG();
	return ret;
}




struct dlm_node_iter
{
	unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
	int curnode;
};


enum {
	DLM_MASTER_REQUEST_MSG    = 500,
	DLM_UNUSED_MSG1,         /* 501 */
	DLM_ASSERT_MASTER_MSG,	 /* 502 */
	DLM_CREATE_LOCK_MSG,	 /* 503 */
	DLM_CONVERT_LOCK_MSG,	 /* 504 */
	DLM_PROXY_AST_MSG,	 /* 505 */
	DLM_UNLOCK_LOCK_MSG,	 /* 506 */
	DLM_UNUSED_MSG2,	 /* 507 */
	DLM_MIGRATE_REQUEST_MSG, /* 508 */
	DLM_MIG_LOCKRES_MSG, 	 /* 509 */
	DLM_QUERY_JOIN_MSG,	 /* 510 */
	DLM_ASSERT_JOINED_MSG,	 /* 511 */
	DLM_CANCEL_JOIN_MSG,	 /* 512 */
	DLM_EXIT_DOMAIN_MSG,	 /* 513 */
	DLM_MASTER_REQUERY_MSG,	 /* 514 */
	DLM_LOCK_REQUEST_MSG,	 /* 515 */
	DLM_RECO_DATA_DONE_MSG,	 /* 516 */
	DLM_BEGIN_RECO_MSG,	 /* 517 */
	DLM_FINALIZE_RECO_MSG	 /* 518 */
};

struct dlm_reco_node_data
{
	int state;
	u8 node_num;
	struct list_head list;
};

enum {
	DLM_RECO_NODE_DATA_DEAD = -1,
	DLM_RECO_NODE_DATA_INIT = 0,
	DLM_RECO_NODE_DATA_REQUESTING,
	DLM_RECO_NODE_DATA_REQUESTED,
	DLM_RECO_NODE_DATA_RECEIVING,
	DLM_RECO_NODE_DATA_DONE,
	DLM_RECO_NODE_DATA_FINALIZE_SENT,
};


enum {
	DLM_MASTER_RESP_NO = 0,
	DLM_MASTER_RESP_YES,
	DLM_MASTER_RESP_MAYBE,
	DLM_MASTER_RESP_ERROR
};


struct dlm_master_request
{
	u8 node_idx;
	u8 namelen;
	__be16 pad1;
	__be32 flags;

	u8 name[O2NM_MAX_NAME_LEN];
};

#define DLM_ASSERT_MASTER_MLE_CLEANUP      0x00000001
#define DLM_ASSERT_MASTER_REQUERY          0x00000002
#define DLM_ASSERT_MASTER_FINISH_MIGRATION 0x00000004
struct dlm_assert_master
{
	u8 node_idx;
	u8 namelen;
	__be16 pad1;
	__be32 flags;

	u8 name[O2NM_MAX_NAME_LEN];
};

struct dlm_migrate_request
{
	u8 master;
	u8 new_master;
	u8 namelen;
	u8 pad1;
	__be32 pad2;
	u8 name[O2NM_MAX_NAME_LEN];
};

struct dlm_master_requery
{
	u8 pad1;
	u8 pad2;
	u8 node_idx;
	u8 namelen;
	__be32 pad3;
	u8 name[O2NM_MAX_NAME_LEN];
};

#define DLM_MRES_RECOVERY   0x01
#define DLM_MRES_MIGRATION  0x02
#define DLM_MRES_ALL_DONE   0x04

/*
 * We would like to get one whole lockres into a single network
 * message whenever possible.  Generally speaking, there will be
 * at most one dlm_lock on a lockres for each node in the cluster,
 * plus (infrequently) any additional locks coming in from userdlm.
 *
 * struct _dlm_lockres_page
 * {
 * 	dlm_migratable_lockres mres;
 * 	dlm_migratable_lock ml[DLM_MAX_MIGRATABLE_LOCKS];
 * 	u8 pad[DLM_MIG_LOCKRES_RESERVED];
 * };
 *
 * from ../cluster/tcp.h
 *    NET_MAX_PAYLOAD_BYTES  (4096 - sizeof(net_msg))
 *    (roughly 4080 bytes)
 * and sizeof(dlm_migratable_lockres) = 112 bytes
 * and sizeof(dlm_migratable_lock) = 16 bytes
 *
 * Choosing DLM_MAX_MIGRATABLE_LOCKS=240 and
 * DLM_MIG_LOCKRES_RESERVED=128 means we have this:
 *